Knowledge Management

The Knowledge page lets you build and manage your organization's AI-powered knowledge base. Ingest documents into vector indices, organize content with categories, monitor vector endpoints, and run semantic search across your indexed data — all from a single interface.

Getting Started

Open Knowledge from the sidebar. The page requires the knowledge_query, knowledge_ingest, or knowledge_endpoint capability on your subscription and the KNOWLEDGE_READ permission on your role. You also need an active vector database resource provisioned in your infrastructure stack.

If the vector database is not yet provisioned, the page displays a setup prompt. Contact your administrator to enable the resource through the admin portal's infrastructure management section.

Interface Overview

The Knowledge page is organized into tabbed sections. The tabs you see depend on your role and capabilities:

Tab	Description	Visibility
Data Management	Ingest files, manage embedded file records, and organize categories	All knowledge users
Infrastructure	View vector endpoints and manage vector indices	Tenant admins (hidden for developer accounts)
Search & Query	Interactive semantic search playground	Superuser access only
AI Chat	RAG-powered conversational search	Superuser access only

Developer accounts see only the Data Management tab with access restricted to their personal index.

Data Management

The Data Management tab is where most day-to-day knowledge base work happens. It contains two main sections: the ingestion wizard and the embedded files manager.

Data Ingestion

The ingestion wizard walks you through adding files to the knowledge base in a multi-step process:

1. Select source — choose between uploading new files from your device or selecting existing files from platform storage
2. Choose files — browse and select one or more files to ingest. Use the search and pagination controls to find files across large libraries
3. Select operation — choose to ingest new files, update existing records, or delete records from the index
4. Choose target index — pick the vector index where the files will be embedded and stored
5. Assign categories — configure data categories for access control. You can replace all existing categories, add new ones, or remove specific ones
6. Review and execute — review the summary and confirm. The platform processes the files in the background using your configured execution strategy

A progress indicator tracks each file as it moves through text extraction, chunking, embedding, and indexing.

Tip: Assign categories during ingestion to control which users and chat personas can access the indexed content later.

Embedded Files

The embedded files manager displays a data grid of all files that have been ingested into the knowledge base:

Column	Description
File Name	Name of the ingested file
Index	The vector index containing this file's embeddings
Categories	Data categories assigned to the record
Status	Processing state: Processing, Completed, or Failed
Chunk Count	Number of text chunks created from the file
Embedding Model	The AI model used to generate embeddings
Error	Error message if processing failed

Use the toolbar to search, filter by index or category, and perform bulk operations.

Bulk Actions

Select files with checkboxes to access bulk operations:

• Update categories — add or remove category assignments across multiple files
• Toggle inference — enable or disable files for AI inference (controls whether the AI Chat can reference them)
• Sync — reconcile database records with the vector store to fix any mismatches
• Delete — remove selected file records and their vectors from the index

Status Reference

Status	Meaning
Processing	The file is being chunked, embedded, and indexed
Completed	The file is fully indexed and available for search
Failed	Processing encountered an error — check the error column for details

Infrastructure

The Infrastructure tab gives administrators visibility into the vector database connections and indices that power the knowledge base.

Vector Endpoints

Vector endpoints represent connections to your cloud vector database service (such as Azure AI Search). Each endpoint is tied to an infrastructure stack and displays:

• Provider — the cloud vector database provider
• Status — connection health
• Endpoint URI — the service URL
• Dimensions — the embedding vector size

Use Sync from Stack to refresh endpoint information from your infrastructure resources. Click Test Connection on any endpoint to verify connectivity.

Note: Vector endpoints are provisioned through your infrastructure stack. You cannot create or delete them directly from this page.

Vector Indices

Vector indices are the logical collections where document embeddings are stored. Each index is associated with a vector endpoint and configured with:

• Name — a descriptive identifier for the index
• Embedding model — the AI model used to generate vectors (e.g., text-embedding-ada-002)
• Chunk size — how many tokens each text chunk contains
• Categories — which data categories are associated with this index
• File count — number of ingested file records

Managing Indices

• Create — click Create Index to set up a new index. Choose a name, embedding model, and initial categories. The platform creates the corresponding collection in the cloud vector store
• Edit — modify the index name, description, or category assignments. Changing the embedding model on an existing index requires re-ingesting all files
• Delete — remove the index and all its vectors from the cloud store. This action is irreversible
• Refresh — update the index metadata from the vector store
• Test Search — run a quick test query against the index to verify it returns results

Important: Deleting a vector index permanently removes all embedded content from that index. File records referencing the deleted index will be cleaned up automatically.

Search & Query

The Search & Query tab provides an interactive playground for testing semantic search against your knowledge base. This tab is available to superusers only.

Running a Search

1. Enter a natural language query in the search box
2. Configure search parameters:
   • Top K — maximum number of results to return
   • Score Threshold — minimum similarity score (0–1) to include a result
   • Indices — select one or more vector indices to search across
   • Categories — filter results by data category
   • Endpoints — choose which vector endpoint to query
3. Click Search to execute the query

Results display with similarity scores, source file information, and matching text snippets. Higher scores indicate stronger semantic matches.

Search History and Saved Searches

The playground maintains a search history for your session. You can also save frequently used queries with their parameter configurations for quick re-use.

Index Analytics

From the search interface you can also access index-level analytics:

• Test Index — verify that a specific index is responding correctly
• Index Stats — view document counts, category breakdowns, and search timing metrics

AI Chat (RAG)

The AI Chat tab provides a conversational interface for querying the knowledge base using retrieval-augmented generation (RAG). This tab is available to superusers only.

How It Works

Type a question and the system retrieves relevant documents from the knowledge base, then uses them as context for generating an AI response. The conversation maintains context across multiple turns.

Chat Filters

Configure these parameters to control which knowledge is included:

• Categories — restrict retrieval to specific data categories
• Index — select which vector index to query
• Score Threshold — minimum similarity score for retrieved documents
• Max Documents — limit the number of documents used as context

Responses include citations linking back to the source documents.

Tip: Start with broad filters and narrow down if results are too noisy. A lower score threshold returns more results but may include less relevant matches.

Categories

Data categories control access to knowledge base content across the platform. Categories are shared between the Knowledge, Files, and Documents modules and stay synchronized automatically.

Managing Categories

1. Open the category manager from the Data Management tab or the Knowledge Hub settings
2. View existing categories with their display names, descriptions, and visibility settings
3. Create, edit, or delete categories as needed

Category Properties

Property	Description
Slug	Unique identifier (auto-generated from the display name)
Display Name	Human-readable label shown in the UI
Description	Optional explanation of what the category covers
Public	Whether the category is accessible to unauthenticated users (e.g., public-facing bots)
Default Roles	Which user roles have access to content in this category by default

How Categories Affect Search

When a user or chat persona performs a knowledge search, the platform filters results based on the categories they have access to. Only documents tagged with matching categories appear in search results.

Categories assigned to files in the Files page automatically propagate to their knowledge base records, and changes made in either direction stay in sync.

Knowledge Hub Integration

The Knowledge Hub is a dashboard tab (available in some product configurations) that provides a streamlined interface to the knowledge base for everyday users. It reuses knowledge categories and the document library without exposing the full infrastructure management tools.

The Knowledge Hub offers two subtabs:

• Search & Discover — semantic search across the knowledge base with category filters and result previews
• Document Library — browse and manage documents with collection-based organization, powered by the same category system

For detailed information about these features, see the Knowledge Hub in-app help topics.

Best Practices

• Organize with categories — create categories that reflect your team's access needs (e.g., by department, project, or sensitivity level)
• Choose embedding models carefully — different models produce different quality results. Test with a sample dataset before ingesting your full library
• Set appropriate chunk sizes — smaller chunks give more precise search results but increase storage; larger chunks provide more context per result
• Use inference toggles — disable inference on files that should remain indexed but not surfaced in AI Chat responses
• Monitor index health — periodically check index stats and run test searches to verify quality
• Review category sync — if you manage categories in both the Files and Knowledge pages, verify they stay consistent using the sync tools

Troubleshooting

Vector database not available

Problem: The Knowledge page shows a message that no vector database is provisioned. Solution: Contact your administrator to provision a vector database resource in the infrastructure stack. The knowledge plugin requires an active vector DB endpoint to function.

Ingestion fails for certain files

Problem: Some files show "Failed" status after ingestion. Solution: Verify that the file contains extractable text. Binary-only formats (images without OCR, compiled binaries) cannot be meaningfully embedded. Check the error column for specific failure details.

Search returns no results

Problem: A semantic search query returns zero matches. Solution: Verify that files have been ingested into the selected index and that their status is "Completed." Check that the selected categories match the categories assigned to the ingested files. Try lowering the score threshold or increasing the top K value.

Category changes not reflected

Problem: You updated a category in Files but the change does not appear in Knowledge. Solution: Category synchronization happens automatically via platform signals, but there may be a short delay. If changes do not propagate, use the Sync tool in the embedded files manager to reconcile records manually.

Developer account limitations

Problem: You cannot see the Infrastructure, Search, or AI Chat tabs. Solution: Developer accounts are limited to the Data Management tab with access to a personal vector index only. Contact your administrator for elevated access.

Related Topics

• Files Management — User Guide
• AI Chat — User Guide
• Dashboard Navigation Guide

---

⏱️ Read time: 15 minutes | 📊 Difficulty: intermediate | 🔄 Last updated: 2026-03-30